On November 14 the FDA will decide whether to approve ivonescimab, a lung-cancer drug that Summit Therapeutics paid the Chinese biotech Akeso $500 million upfront to bring to America. When the agency’s reviewers sit down with the file, they will be staring at two lines on a graph that tell two different stories. One line, the one Summit likes to put on a slide, pulled apart cleanly. The other, the one that tracks whether patients actually lived longer, barely came apart at all.
It happens that a medical newsletter chose this same season to repost, for free, a lecture on how to read a Kaplan-Meier curve, the staircase-shaped line that sits at the center of how cancer trials get read and how cancer drugs get approved. The lecture is a dry thing taught to first-year trainees. Ivonescimab is very nearly its worked example, because the whole case turns on the one lesson it exists to teach: when a trial hands you two curves, learn which one to believe.
Ivonescimab is essentially the whole of Summit Therapeutics. The company licensed it from Akeso in early 2023, took the rights to develop it in the United States, Canada, Europe, and Japan, and built itself around it. It is a bispecific antibody, engineered to block PD-1 and VEGF at once, and the pitch behind it is genuinely big: in a separate Akeso-run trial in China, ivonescimab went head to head against Merck’s Keytruda, the best-selling drug in the world, and beat it on tumor progression. That result is why anyone outside oncology has heard of Summit at all. HARMONi was supposed to be the Western confirmation.
Two curves, two stories
HARMONi enrolled 438 patients with advanced EGFR-mutated non-small-cell lung cancer whose disease had already outrun a modern targeted pill, and split them evenly between ivonescimab plus chemotherapy and placebo plus chemotherapy, double-blind, across 114 centers on three continents. These were patients out of easy options. The trial asked two questions, and it answered them differently.
The first was progression-free survival: the time before a scan shows the tumor growing again. Here ivonescimab won cleanly, a median of 6.8 months against 4.4, a hazard ratio of 0.52 with a 95 percent confidence interval of 0.41 to 0.66, the kind of separation that clears significance with room to spare. Call it ten extra weeks before the cancer stirred on imaging.
The second was whether patients lived longer. Median overall survival came in at 16.8 months with the drug against 14.0 without it, a hazard ratio of 0.79 with a 95 percent confidence interval running from 0.62 to 1.01. That upper bound sits just past 1.0, which means the survival benefit never cleared the trial’s own bar for significance; the reported p-value, 0.057, missed by a hair. The progression curve pulled apart. The survival curve never did.
The curve patients feel
This is the exact gap the Kaplan-Meier lecture teaches you to look for, and the one Sensible Medicine has spent years warning oncologists about. Progression-free survival, as its writers put it, can be “an endpoint that is invisible to patients, maybe only visible to radiologists.” A tumor that is technically not growing on a CT scan does not announce itself to the person carrying it. Living longer does. Feeling better does. Progression-free survival is neither.
The cautionary case those writers keep returning to is COSMIC-313, a kidney-cancer trial where a three-drug regimen improved progression-free survival and then delivered no overall-survival benefit at all, median survival landing at roughly 42 months in both arms while piling on toxicity and more discontinuations. Patients spent that borrowed radiographic time on a harsher drug, not on more life. Ivonescimab is not COSMIC-313, and its survival trend at least points the right way. Summit argues, with some reason, that the follow-up is still maturing: in the Western subgroup the death-risk hazard ratio has drifted from 0.98 at the first analysis toward 0.76 in the latest cut.
But the company’s public framing leans on the curve that separated, and a reader trained on that lecture is entitled to ask the older question first. At this data cut, does the drug help anyone live longer? The trial’s own answer is: not yet proven.
What the delay costs
There is also a bill that never shows up on the progression curve. Serious treatment-related adverse events struck 28 percent of patients on ivonescimab against 15 percent on chemotherapy alone, close to double. Grade 3-4 drops in neutrophils, white cells, and platelets all ran higher with the drug. Four patients in the ivonescimab arm died of treatment-related causes, against five in the placebo arm, a wash on the worst outcome, and a reminder that the extra months before progression were not handed out for free.
None of this makes ivonescimab a bad drug. For someone who has burned through the targeted options, ten more weeks before the scan turns is not nothing, and a bispecific that can stand toe to toe with Keytruda is no small piece of engineering. It makes it an unfinished one. On November 14 the FDA will be looking at the same two curves everyone else has seen, and it will have to decide, the way every oncologist downstream will, which of them it is buying.
Sources
- The Lancet Oncology – Le et al., HARMONi phase 3 trial of ivonescimab in EGFR-mutated NSCLC (2026)
- FiercePharma – Summit updates ivonescimab survival data ahead of FDA decision date
- The ASCO Post – HARMONi-2: Ivonescimab Outperforms Pembrolizumab as First-Line Treatment in NSCLC
- ASCO Daily News – COSMIC-313 Final Results: No OS Improvement With First-Line Cabozantinib/Nivolumab/Ipilimumab
- Sensible Medicine – “The Illusion of Progression-Free Survival”
- Sensible Medicine – “Lecture 6: KM Curves”
- BusinessWire – Summit Therapeutics closes $500 million in-license deal with Akeso for ivonescimab